Arguments
New study questions the accuracy of satellite atmospheric temperature estimates
Posted on 7 November 2014 by John Abraham
Over the past decades, scientists have made many measurements across the globe to characterize how fast the Earth is warming. It may seem trivial, but taking the Earth’s temperature is not very straightforward. You could use temperature thermometers at weather stations that are spread across the globe. Measurements can be taken daily and information sent to central repositories where some average is determined.
A downside of thermometers is that they do not cover the entire planet – large polar regions, oceans, and areas in the developing world have no or very few measurements. Another problem is that they may change over time. Perhaps the thermometers are replaced or moved, or perhaps the landscape around the thermometers changes which could impact the reading. And of course, measurements of the ocean regions are a whole other story.
An alternative technique is to use satellites to extract temperatures from radiative emission at microwave frequencies from oxygen in the atmosphere. Satellites can cover the entire globe and thereby avoid the problem with discrete sensors. However, satellites also change over time, their orbit can change, or their detection devices can also change.
Another issue with satellites is that the measurements are made throughout the atmosphere that can contain contaminants to corrupt the measurement. For instance, it is possible that water droplets (either in clouds or precipitation) can influence the temperature readings.
So, it is clear that there are strengths and weaknesses to any temperature measurement method. You would hope that either method would tell a similar story, and they do to some extent, but there are key differences. Basically, the lower atmosphere (troposphere) is heating slower than the Earth surface.
In fact, for the time period 1987–2006, the temperatures among the four groups that collect satellite data ranges from 0.086°C per decade to 0.22°C per decade. In more recent years, the trend is much reduced, and for two of the leading satellite groups (University of Alabama at Huntsville and Remote Sensing Systems), temperatures are basically flat.
The recent flatness in satellite temperatures as surface temperatures continue to rise has presented a quandary for scientists. Are both results real? Is there some reason they diverge? Is one measurement more accurate than the other? This is one of the areas of very active research.
A contribution to this question appeared last week by researcher Fuzhong Weng and his colleagues. The paper, published in Climate Dynamics, claimed to find the reason for much of that difference – the authors report that the satellite trends could be off (too cold) by perhaps 30%. If true, this work would go a long way toward reconciling the differences between surface and satellite measurements.
While this paper is getting a lot of attention, I am suggesting a more cautious approach. There are a number of issues and questions which must be answered before we can close the books on this issue and the paper has received some critical attention from other scientists. Before we get into that, let’s talk about what the study found and how they made their discovery.
For a few decades, satellites have measured radiant emission from oxygen in the atmosphere and have related these measurements to temperatures. As satellites orbit the Earth, the microwave instrument on-board scans the atmosphere below them every 8 seconds or so and scientists apply what are called weighting functions to extract information from different altitudes. Each of the microwave “channels” uses a different weighting function so as to obtain information at different heights. The four channels most associated with atmospheric temperatures are Advanced Microwave Sounding Unit channels 3, 5, 7, and 9 in the current fleet of satellites.
The radiant emission received by the satellite can be influenced by other components in the atmosphere, in particular cloud liquid water. Many years ago, the impact of cloud liquid water was considered and various attempts were made to eliminate its influence through a filtering process. It is well known that cloud liquid water can influence the measurements, the real question is by how much?
The vigorous debate from the 1990s has been rekindled in the present Weng study. This new work segregates the Earth system by levels of cloudiness and precipitation in the atmosphere. The authors term “clear-sky” conditions corresponding to less than 10 grams of water per square meter of surface area. The authors then envisioned a cloud layer atop a raining region whose total height extends approximately 4 km vertically from the Earth surface.
In their analysis, they considered different droplet sizes ranging from .05 mm to 1 mm. Finally, the impact on the satellite channels (AMSU-A channels 3, 5, 7, and 9) was determined. It was found that the lowest channel (channel 3 which is primarily focused on the near surface region) was significantly impacted by the presence of cloud liquid water.
As you move higher into the atmosphere, the impact on temperatures was much reduced. When you look at the trends (change in temperatures with time), the two lowest elevation channels are higher when the impacts of clouds are removed. What this means is, measurements made in cloudy skies gives a lower warming trend of the atmosphere.
The authors state,
A decrease in brightness temperature can be associated with cloud and precipitation scattering, rather than physical temperature in the lower and middle troposphere and therefore, trends from microwave sounding data could be misleading if the brightness temperature from all weather conditions are averaged as representative of atmospheric physical temperature.
The trend calculated from the clear-sky data is thus not only larger but also more reliable ... It is shown that the atmospheric warming trends in the middle latitudes are significantly larger when cloud effects are removed … The scattering and emission effect of clouds and precipitation significantly reduces the values of the warm trends in the low and middle troposphere derived from microwave data.
Simply put, when you eliminate the effect of clouds, the atmosphere is warming faster than we thought and the divergence between land thermometers and satellites largely disappears.
Of course, whenever a study that is this significant is published, there is deserved skepticism. We have to be guarded in our acceptance until further work is done and until other teams have had a chance to review the findings. I asked others who work in this area to find their impressions.
Dr Abraham:
The version of the OP over at The Guardian feels a bit incomplete, as if there were more people whose impressions you had gathered that weren't included, and the impression I have is that there was more to be said about the critical attention Weng et al received, only the post was truncated before you got around to discussing it.
Was that your intent?
Why is it that the Grauniad's only 'expert' commentator is Roy Spencer of the Heartland Institute? He thinks the globe is cooling and CO2 is good for the economy. His 2011 paper on climate sensitivity was so bad that the editor who published it called it "fundamentally flawed" and resigned. It's like asking Bernie Madoff to comment on legislation to prevent tax fraud.
Ricardo K @2, as one of the two original developers of the satellite temperature series, and co-author of the oldest and most cited satellite temperature series, Spencer is well qualified to talk on this issue. This is one of the few occasions where including his views does not constitute false balance (despite his known biases).
Better understanding the differences between the trends of very different evaluations is important. But inconsistencies of trends of short-term averages in the different methods would not prove either is wrong. It may only indicate that different short term factors have a different effect on each measurement.
It could be found that in spite of the eventually understood reasons for the differences both methods produce comparable trends of the averages of longer durations of measurement like a 360 month rolling average. However, the satellite record developed by Spencer is only 36 years long. The Spencer satellite data would only be able to present 6 years of such an average (1979-2008, through 1985-2014). But it is fairly obvious that the trend through those 7 years would be clearly increasing since the 13 month averages on Spencer's graph since 2008 have been warmer than the values before 1985. A simplistic evaluation of this 7 year trend would be:
Therefore, until there is a significantly longer record of satellite data there is little justification for claims that the satellite record is more accurate and that the surface temperature data is incorrect. Also, there is little justification for a claim that we need to wait until that longer satellite record is established.
My mistake in failing to fully revise my presentation to the duration of the trend in the satellite data set being 6 years rather than 7.
"Simply put, when you eliminate the effect of clouds, the atmosphere is warming faster"
....mmm wouldn't you want to be more specific? Say instead
Simply put, when you eliminate the effect of clouds on the satellite readings, the atmosphere is warming faster...
For there is also the effect of clouds on the actual temperature we are attempting to measure. This is a nasty bit of parsing because the effects of clouds appear in so many places and the quote miners are SO ....
The most curious thing about the lower troposphere temperature data from satellites is that there is a large systematic difference between the results given by UAH and by RSS; even though both are using the same raw data. The UAH group (the one Roy Spencer works on) shows much MORE warming. It is the RSS group which shows basically flat.
Has anyone looked into this difference?
The long term trends of the RSS and UAH global temperature records for the TLT altitude is basically the same as the long-term trend in the NOAA and GISS surface temperature records since the mid-1970s.
The 21-year running averages of all 4 data sets give pretty much identical warming trends of 0.14-0.15 C degrees per decade up until 2004. If that trend has continued over the past decade, then all 4 data sets are showing that the 2014 global temperature value will be pretty much on the trend line, which is what you would expect for an ENSO-neutral year.
The only real difference in the data sets is that the satellite monthly temperature data shows an exaggerated escalator effect. This results from the satellite measurements have a much larger (by about a factor of 2) response to El Nino and La Nina variations than the surface temperature data sets.
WRyan - I agree that the majority of the disparity in trends disappears once ENSO is accounted for, which, if the paper is correct, would lead to a rather scary conclusion. The SST records are biased low.
This is not a new idea I might add.
Tristan, I was thinking of that Cowtan and Wray paper you link to also. I suspect you were referring to the fact that Cowtan & Wray used a kriging method (similar to what BEST did over land) to fill in the gaps between measured temperatures and that found that past SST records were biased low.
However, they also attempted to use satellite data to help fill in the gaps from surface measurements. What is interesting in relation to the new paper is that Cowtan & Wray found that combining the satellite and thermometer records worked well over land and ice... but not the oceans. When they tried it with the oceans they got significant mismatches. It would be interesting to see if the adjusted satellite calculations from Weng & co prove a better match if/when run through Cowtan & Wray's methodology... and/or how closely they line up with the results Cowtan & Wray got using kriging on surface readings.
CBDunkerson - The paper you mention is by Cowtan and Way (not Wray). The sheer variety of misspellings of the author names here and there has been amusing, but it's important to give proper credit.
WRyan,
I understand that when you referred to the 21 year averages in the satellite data up to 2004 you meant the mid-point of the latest 21 year period ending in 2014. However, it may be clearer to refer to it as the average of 21 years ending in 2014, especially if you speculate about trend over the 'past decade'. You really are speculating about the trend that will be seen in the data through the 'coming decade' And looking at the trend of averages of shorter durations in the more recent data, as some may be tempted to do to see what is happening ion the 'recent past decade' does not indicate what the longer term averages will be. The data sets are filled with rather random rapid short-term changes.
Even the 21 year average you used to get a reasonable length for the trend in the satellite data may be a short duration. A more rigorous evaluation of the 30 year averages in the satellite data would probably alos indicate a rate cloase to 0.15 C per decade. I tried to make my simple assessment deliberately conservative.
p.s. The preferred standard for establishing regional climate expectations by the WMO member organisations has been the evaluation of the most recent 30 years of observations. However, they have been learning that rapid climate change requires different evaluations. What happened over the past 30 years in any region is no longer as reliable as it used to be for determining what to expect in the near future.